Mapping Permanent irrigated Grasslands Using Sentinel-2 Data based on temporal patterns

Irrigation has a strong impact on water resources as groundwater. Grassland irrigation was often done using flooding technics, which mobilize large amount of water that might have effect on groundwater recharge and discharge. Mapping those irrigated grassland is therefore a crucial information to assess ground-water dynamic. Here we propose a land use classification approach based on the temporal patterns, that are specific to grassland to avoid the use of training data sets. Thanks to the frequent acquisition allowed by recent satellite missions as Sentinel 2, we used time series of leaf area index (LAI) to identify grass cuts. This approach was applied to identify irrigated permanent grasslands in the Crau area (south of France). These are regularly mown with two to four cuts during the May-October period that leads to a specific temporal pattern of LAI. An algorithm was designed to detect the number of cuts in the temporal LAI signal (see Figure 1). The algorithm includes some filtering to remove noise in the signal that might lead to false cut detection. A pixel is considered as a grassland if the number of detected cuts ranges from 2 to 4 while intensive alfalfa sometimes led to 5 cuts. A data set covering five years (2016-2020) was used. The cut number detection was done at the pixel level and then results are aggregated at the field level (120000 fields over the area). A validation data set including 800 fields was used to assess the performances of the classification. We computed the Cohen Kappa index, and obtained results ranging between 0.93-0.99 according to the year (see Table 1). These results are slightly better than other supervised classification methods that include training data sets. Grassland detection obtained with different years was used to evaluate the capacity to detect land use change. Moreover, mowing calendar can be derived and used for farming practices analysis or crop modelling over large areas than can be used to spatialize the groundwater recharge.